[Computer-aided diagnosis of interstitial lung diseases]

Area of Science:

• Radiology
• Medical Imaging Analysis
• Computer-Aided Diagnosis

Background:

• Interstitial lung disease (ILD) diagnosis relies on expert interpretation of chest radiographs.
• Quantitative texture analysis offers potential for objective assessment of lung abnormalities.

Purpose of the Study:

• To develop and evaluate an automated method for quantitative lung texture analysis in digital chest radiographs.
• To detect and characterize interstitial lung disease using computer-aided classification.
• To assess the impact of digital parameters on the accuracy of the automated analysis.

Main Methods:

• Quantitative texture measures (root-mean-square variation, power spectrum moment) were derived from lung textures.
• An automated classification system was developed using these measures and a clinical database.
• The system employs three tests for focal, localized, and diffuse abnormal patterns.
• Receiver operating characteristic (ROC) curves compared computerized performance with radiologists' performance.

Main Results:

• Computerized analysis demonstrated performance comparable to human observers in distinguishing mild interstitial lung disease from normal lungs.
• Digital parameters (pixel size, ROI size, quantization levels) were investigated to optimize the computerized scheme.
• Texture measures from the International Labour Organization (ILO) pneumoconioses standard radiographs correlated with ILO classification categories for small opacities.

Conclusions:

• The developed automated method provides quantitative physical measures of lung textures.
• This computer-aided approach can serve as a valuable tool for radiologists in assessing interstitial lung diseases.
• Optimization of digital parameters is crucial for the effective design and implementation of the computerized scheme.